Do antidepressants work Image: Tom Varco, CC BY-SA 3.0, via Wikimedia Commons

Do antidepressants work

The question in the title has a short answer, which is yes, and a longer one, which is the only useful one. The drugs work better than placebo in every large analysis ever done. The difference is small on average, larger for some patients than others, and was overstated for twenty years by the way the trials were published. All three facts are needed to say anything sensible to a person deciding whether to take one.

This article describes the evidence. It is not medical advice, and nobody should start or stop a prescribed medicine on the strength of a blog post; stopping an antidepressant suddenly is itself harmful, and the decision belongs with the patient and a clinician who knows them.

The largest analysis

In 2018 Andrea Cipriani and colleagues published in The Lancet a network meta-analysis of 522 randomized trials of 21 antidepressants in adults with major depressive disorder, with 116,477 participants in all, including unpublished trials obtained from manufacturers and regulators; the paper is at the journal. Every one of the 21 drugs was more effective than placebo over the eight weeks of a typical trial. The odds of responding, defined as a halving of the symptom score, were between about 1.4 and 2.1 times higher on a drug than on placebo, depending on the drug. Expressed as a standardized difference in symptom scores, the average effect was a Cohen's d of about 0.3.

Two things about the figure. A d of 0.3 is, by the conventions of Unit 1, between small and medium, and it is similar to the effect of many accepted treatments in medicine. And it is an average over people, most of whom improve substantially on placebo as well: in these trials roughly half of the patients on a drug respond and roughly a third to 40% of those on placebo, which means that one additional patient responds for every seven to nine treated. That number, the number needed to treat, is the honest way to say how often the drug makes the difference, and Unit 12 of the masterclass computes it from the response rates.

The file drawer

Why did the field spend so long believing the effect was larger? In 2008 Erick Turner and colleagues compared the trials that drug companies had registered with the United States Food and Drug Administration, which sees every trial, with the trials that had appeared in journals. Of 74 registered trials of 12 antidepressants, 38 had a positive result and 37 of those were published. Of the 36 with a negative or questionable result, 22 were never published and 11 were published in a way that made them look positive. A reader of the literature saw 94% positive trials; the regulator saw 51%. The effect size in the published trials was about a third larger than in the full set. The paper, in the New England Journal of Medicine, is at the journal, and it is one of the clearest demonstrations of publication bias in any field. Trial registration is now required, and the Cipriani analysis, which used the unpublished data, is the corrected picture.

The same year Irving Kirsch and colleagues obtained the FDA data for four of the newer drugs and reported in PLOS Medicine that the drug-placebo difference was below the threshold that the British guideline body then used for clinical significance, except in the most severely depressed patients. The headline that antidepressants do not work came from that paper. The analysis itself said something more specific, which the next section takes up.

Who benefits

The average hides a distribution. Jay Fournier and colleagues pooled individual patient data from six trials in 2010 and found that the benefit over placebo was negligible for patients with mild or moderate symptoms at the start and substantial for those with severe depression; the paper is at JAMA. Later analyses with more data have found a smaller dependence on severity than that, but the direction has held. In 2022 Marc Stone and colleagues at the FDA analysed the individual records of 73,388 patients in 232 trials, free at the BMJ, and found that the average difference between drug and placebo, about 1.75 points on a 52-point symptom scale, was best explained by a model in which about 15% of patients have a large response to the drug that they would not have had to placebo, while the rest respond similarly to both. If that model is right, the drug is of little use to most of the people who take it and of great use to a minority, and nobody yet knows how to tell in advance who is in the minority.

What the trials do not cover

Trials last eight weeks and depression lasts longer. The best evidence on continuing treatment comes from stopping it: in a 2021 trial of 478 patients in English general practice who had recovered on an antidepressant, 56% of those switched to placebo relapsed within a year against 39% of those who continued. The drugs reduce relapse as well as symptoms. Against that stand side effects, which are common and differ by drug, and withdrawal symptoms on stopping, which are more frequent and longer lasting than the trials suggested and are the reason for tapering slowly.

The mechanism is also less settled than the public was told. The idea that depression is a deficiency of serotonin that the drugs correct was always a hypothesis, and a 2022 umbrella review of the evidence for it found little support; the drugs raise serotonin within hours and relieve symptoms, when they do, over weeks, so whatever they do runs through slower changes that are still being worked out. A drug can work without its advertised explanation being right, and these do.

What else works

Psychotherapy, in particular cognitive behavioural therapy, has an effect against control conditions of about the same size as the drugs in meta-analyses by Pim Cuijpers and colleagues, and the combination of the two is better than either alone in the trials that have compared them; a 2020 meta-analysis of that question in World Psychiatry is free to read. Exercise has an effect in trials, smaller and less certain. The choice among them depends on severity, on what is available, on what the person has tried before, and on what they prefer, which is why it belongs to a conversation rather than a table.

Reading the figures for one person

Put the evidence together and the sentence a clinician can honestly say is this: these drugs help more people than placebo does, by a margin that is modest on average and larger for severe depression; most people who improve on them would have improved somewhat anyway; a minority improve a great deal; the drugs reduce the chance of relapse; they have side effects and need to be stopped slowly; and talking therapy works about as well and combines with them. That is less satisfying than either "they work" or "they are placebos with side effects", and it is what the 522 trials say. Reading a trial, a response rate, and a number needed to treat is the skill the masterclass teaches in Unit 12, and this question is the reason it is worth having.

Study the mind properly

The Psychology Masterclass: weekly units, quizzes, a final exam, and a certificate.
About the program